studio-workflow

Studio Workflow

Generate a demo video from a raw Playwright trace recorded by recast-studio.

When to Use

• User recorded a session with recast-studio and wants to generate a video
• User says "generate video from trace", "process my recording", "make a demo from this trace"
• User points to a directory with trace.zip and .webm files

Input

A directory path containing:

• trace.zip — Playwright trace
• *.webm — screen recording

Produced by: npx recast-studio <url>

Workflow

Execute these steps in order. Each step produces artifacts the next step needs.

Step 1: Parse the trace

Read the trace directory and parse trace.zip to extract all actions. Use the playwright-recast parseTrace function:

const { parseTrace } = await import('<recast-root>/src/parse/trace-parser.js')
const trace = await parseTrace('<input-dir>/trace.zip')

List all actions with their index, method, selector/URL, value, and timestamp. Mask any values from password-like selectors (containing "password", "passwd", "pwd", "secret").

Present the action list to yourself for analysis.

Step 2: Analyze and group actions

Group the actions into logical steps. For each step, decide:

Hidden steps — setup that should not appear in the video:

• Navigation to the starting URL (goto)
• Login flows (fill username + fill password + click submit)
• Cookie consent, onboarding dialogs
• Any action the viewer doesn't need to see

Visible steps — the actual demo content:

• The core user journey being demonstrated
• Each visible step groups 1-3 related actions (e.g., click input + fill value = one step)

Track which action indices belong to each step and whether it's hidden.

Step 3: Write voiceover

For each visible step, write 1-2 sentences of voiceover text. Follow the script-writer skill guidelines:

• Narrative arc — always follow: Hook (problem) → Solution/Walkthrough → Result
• Marketing tone — benefit-focused, professional, concise
• Language — infer from the UI language, user's instructions, or conversation context
• Never describe mechanical clicks — describe what the user ACHIEVES
• Never mention credentials or passwords
• Each sentence should be natural for TTS (ElevenLabs)
• First visible step = hook (name the problem being solved)
• Middle steps = solution/walkthrough (what each action achieves for the user)
• Last visible step = result + soft CTA

Step 4: Generate SRT file

Create an SRT file mapping voiceover text to trace timestamps.

Timing rules:

• Each entry starts at the first action's startTime in that step
• Each entry ends at the start of the next visible step
• Last entry ends at its start + 5000ms
• Times are trace-relative (milliseconds from trace start)
• Use standard SRT format: HH:MM:SS,mmm --> HH:MM:SS,mmm

SRT numbering: sequential starting from 1, only visible steps with voiceover.

Write the SRT to <input-dir>/subtitles.srt.

Step 5: Build and run recast pipeline

Create a TypeScript script that runs the recast pipeline. Key decisions:

hideSteps predicate: Build a Set<number> of hidden action indices. The predicate matches actions by their timestamp against the original trace actions.

Speed: duringIdle: 3.0, duringUserAction: 1.0, duringNetworkWait: 2.0

Visual effects: autoZoom (inputLevel: 1.2), cursorOverlay, clickEffect with sound.

Voiceover: ElevenLabsProvider with eleven_multilingual_v2 model. Language code from the voiceover text language. Requires ELEVENLABS_API_KEY.

Subtitles: burn into video with styled ASS subtitles (Arial, bold, white background).

Intro/outro: if user provides paths, add .intro() and .outro() to the pipeline.

Output: <input-dir>/demo.mp4 (or user-specified path).

import { Recast, ElevenLabsProvider } from 'playwright-recast'

const hiddenActionIndices = new Set([/* from step 2 */])

let pipeline = Recast.from('<input-dir>')
  .parse()
  .hideSteps((action) => {
    // Match hidden actions by timestamp
    const idx = traceActions.findIndex(
      a => a.startTime === action.startTime && a.endTime === action.endTime
    )
    return hiddenActionIndices.has(idx)
  })
  .speedUp({ duringIdle: 3.0, duringUserAction: 1.0, duringNetworkWait: 2.0 })
  .subtitlesFromSrt('<input-dir>/subtitles.srt')
  .textProcessing({ builtins: true })
  .autoZoom({ inputLevel: 1.2, clickLevel: 1.0, centerBias: 0.3 })
  .cursorOverlay()
  .clickEffect({ sound: true })
  .voiceover(ElevenLabsProvider({
    modelId: 'eleven_multilingual_v2',
    languageCode: '<detected-language>',
  }))
  .render({
    format: 'mp4',
    resolution: '4k',
    fps: 120,
    burnSubtitles: true,
    subtitleStyle: {
      fontFamily: 'Arial',
      fontSize: 96,
      primaryColor: '#1a1a1a',
      backgroundColor: '#FFFFFF',
      backgroundOpacity: 0.75,
      padding: 40,
      bold: true,
      position: 'bottom',
      marginVertical: 100,
      marginHorizontal: 200,
      chunkOptions: { maxCharsPerLine: 55 },
    },
  })

await pipeline.toFile('<input-dir>/demo.mp4')

Step 6: Report result

After the pipeline completes, report:

• Video path and file size
• Number of visible / hidden steps
• Voiceover texts used
• Duration of output video

User Interaction

• If the trace has fewer than 3 actions, warn and ask if they want to proceed
• If unsure about which actions are setup vs. demo, ASK the user before proceeding
• If user wants different voiceover tone/language, regenerate SRT and re-run pipeline
• Support --no-voiceover — skip TTS, keep subtitles only
• If ELEVENLABS_API_KEY is not set, inform user and suggest subtitle-only mode

Example Session

User: /studio-workflow .recast-studio/

Agent: Reading trace from .recast-studio/trace.zip...

Found 9 actions:
  0: goto https://app.codexis.cz
  1: click "Přihlásit se"
  2: fill #username → "jana@..."
  3: fill #password → "***"
  4: click "Přihlásit"
  5: click .search-input
  6: fill .search-input → "pracovní právo"
  7: click .result-item
  8: click "Stáhnout"

Grouping into steps:
  [hidden] Actions 0-4: Login and navigation
  [visible] Actions 5-6: "Do vyhledávače zadáme klíčový pojem z oblasti pracovního práva."
  [visible] Action 7: "Z výsledků vybereme nejrelevantnější dokument."
  [visible] Action 8: "Jedním kliknutím dokument stáhneme pro offline práci."

Writing SRT to .recast-studio/subtitles.srt...
Running recast pipeline...
  ✔ Speed processing (38s → 18s)
  ✔ Voiceover (ElevenLabs, 3 segments)
  ✔ Click effects + cursor + auto-zoom
  ✔ Rendering 4K

✅ .recast-studio/demo.mp4 (8.2 MB)